The use of probabilistic decision models in technology assessment : the case of total hip replacement

There is increasing recognition that decision modelling is central to health technology assessment and, in particular, to analyses to support formal decision making regarding the funding of the use of new technologies. In part, the key role of decision analysis stems from the need to handle multiple sources of uncertainty in the available evidence. The use of probabilistic decision analysis is a means of reflecting the parameter uncertainty in models and presenting this in a comprehensible manner to decision makers. In this article, we demonstrate the potential role of probabilistic models using the case study of total hip replacement surgery.A cost-effectiveness model was constructed to compare the Charnley and Spectron hip prostheses in terms of lifetime costs and quality-adjusted life-years (QALYs). Revision rates were estimated from the Swedish National Total Hip Arthroplasty Register (1992-2000); the risk of revision with the Spectron prosthesis relative to the Charnley prosthesis was 0.67 (95% confidence interval [CI] 0.32, 1.02) for early revisions and 0.26 (95% CI 0.07, 0.46) for late revisions. This lower revision risk resulted in the Spectron generating more QALYs than the Charnley prosthesis. Based on mean costs and QALYs, the Spectron results in cost savings in younger patients, and generates incremental cost-effectiveness ratios of between pound1000 and pound16 000 in older patient groups. The probabilistic results from the model indicated that, if it is assumed that decision makers are willing to pay up to pound20 000 per additional QALY, the probability of the Spectron being the more cost-effective prosthesis ranged between 70% and 100%, depending on the age and sex of the patient.This article looks at the application of probabilistic decision modelling using total hip replacement as a case study to emphasis the need for decision models to quantify all sources of parameter uncertainty and to clearly distinguish parameter uncertainty from subgroup heterogeneity.